Sr. Staff Software Engineer @ Writesonic | AI Chatbots | Cybersecurity

Ashish Bedi • Peerlist

Aim is to present a novel approach for object identification, annotation and tracking in a drone video. It involves image segmentation with deep learning, use of CNN, R-CNN.

Multiple object Classification, Annotation and Tracking from

Developed a spatio-temporal Customer Perception Business Intelligence Dashboard for small and medium scale businesses, ranging in multiple verticals, to monitor and analyse customer's perception (by doing sentiment analysis on data obtained from customer reviews and social media) about one's business as gathered from online reviews and customer generated content.

Customer Perception Business Intelligence Dasboard

Implemented a system and an interactive NLP based chat-bot, which provides wardrobe suggestions (based on raw user input query) and enables user to virtually try the wardrobe on their pictures before actually making a purchase.

Virtual Wardrobe- Flipkart Hackday 7

Mobile phones penetration in Rural India is increasing comparable to TV & Radios, which were the traditional source of entertainment and educative information. Although mobile phones can be a great medium for serving the same, but in many rural areas, mobile phones have limited (part of the day) or no data-access. We have attempted to build a system to make the "relevant" or "popular" content available on these phones offline by pushing the data when the phones come online.


Net Connect (Android App)

- Automated  the  mystery  shopping  process,  which  was  earlier  being  done  manually,  to  identify and shop for listings corresponding to probable fraudulent sellers.  Also, took buyer side (buyer authenticity and buyer-return comments) into consideration.
- Implemented an end-to-end system, from obtaining the data present in the form of raw HIVE tables inside BigFoot (Flipkart's internal Data Store) to making the generated reports of probable fraudulent sellers, downloadable from Ops UI (Operations Team), after running them through an ML model for testing seller on different features. Also integrated this implementation with Order Management System to automate ordering of the listings generated in the reports.

Mystery  Shopping  for  probable  fraudulent  sellers  (Data

- Aim was to detect an `uncalled' situation (such as an earthquake or a bomb blast) and the corresponding location, by collecting and analyzing relevant twitter feeds in real-time.
- This was achieved by keyword-based tweet filtering and making the tweets pass through the stages of Part-of-Speech and Named Entity Relation, using Natural Language Processing tools and Stanford Core NLP. Also, geocoded each location by sending a get request to OpenMap API to find coordinates.

Detecting an `uncalled' situation X via Map-Reduce : Big Dat

Added various scheduling policies - FIFO, LRU, Priority Scheduling, SRT, SRJF in xv6 scheduler (using 2 cores).

Added various scheduling policies to xv6 scheduler.

Developed a uni-processor process scheduling simulator, which works in the similar way as the process scheduler works in an OS. Implemented the entire strategy and compared the outcomes of the processes (or jobs) fed to various scheduling strategies like : Round Robin, Shortest Job Remaining First, First Come First Serve.

Uni-processor Process Scheduling Simulator

Worked for the creation of an intelligent assistant for E-Commerce, an AI powered Automated messaging platform that can process the customer orders. Implemented a chat-based personal shopping assistant.
Major technical works I contributed towards: NLP(Natural Language Processing),Web
Crawling, Data Scrapping, Data Normalizing.
Platform(or Toolkit) Used: Stanford core NLP(Natural Language Processing).

Artificial Intelligence assisted buying platform

Developed an autonomous floor cleaning robot hardware prototype which can sweep the floor whilst avoiding obstacles. Developed the algorithm for the same.